Resolving Deadlock Propagation in Distributed Microservice Orchestration Under Byzantine Fault Conditions via Adaptive Consensus Partitioning

Abstract

Deadlock propagation in distributed microservice architectures operating under Byzantine fault conditions represents a critical reliability challenge for large-scale cloud-native systems. Existing consensus protocols exhibit significant latency degradation and livelock susceptibility when node failures exhibit non-crash, arbitrary behavior. This paper introduces Adaptive Consensus Partitioning (ACP), a novel fault-tolerant orchestration framework that dynamically restructures service dependency graphs to isolate Byzantine partitions before deadlock cascades. ACP integrates a lightweight distributed deadlock detection algorithm with a quorum-adaptive voting mechanism, reducing mean time-to-recovery by 61.4% across heterogeneous cluster configurations. Empirical evaluation on a 256-node testbed demonstrates that ACP sustains throughput degradation below 8% under simultaneous 30% Byzantine node injection, outperforming state-of-the-art baselines including BFT-SMaRt and HotStuff variants. These results confirm ACP's viability for production-grade, fault-resilient microservice deployments.